On 09.12.20 22:43, Mike Rapoport wrote:
> From: Mike Rapoport <r...@linux.ibm.com>
>
> memblock does not require that the reserved memory ranges will be a subset
> of memblock.memory.
>
> As the result there maybe reserved pages that are not in the range of any
> zone or node because zone and node boundaries are detected based on
> memblock.memory and pages that only present in memblock.reserved are not
> taken into account during zone/node size detection.
>
> Make sure that all ranges in memblock.reserved are added to memblock.memory
> before calculating node and zone boundaries.
>
> Fixes: 73a6e474cb37 ("mm: memmap_init: iterate over memblock regions rather
> that check each PFN")
> Reported-by: Andrea Arcangeli <aarca...@redhat.com>
> Signed-off-by: Mike Rapoport <r...@linux.ibm.com>
> ---
> include/linux/memblock.h | 1 +
> mm/memblock.c | 24 ++++++++++++++++++++++++
> mm/page_alloc.c | 7 +++++++
> 3 files changed, 32 insertions(+)
>
> diff --git a/include/linux/memblock.h b/include/linux/memblock.h
> index ef131255cedc..e64dae2dd1ce 100644
> --- a/include/linux/memblock.h
> +++ b/include/linux/memblock.h
> @@ -120,6 +120,7 @@ int memblock_clear_nomap(phys_addr_t base, phys_addr_t
> size);
> unsigned long memblock_free_all(void);
> void reset_node_managed_pages(pg_data_t *pgdat);
> void reset_all_zones_managed_pages(void);
> +void memblock_enforce_memory_reserved_overlap(void);
>
> /* Low level functions */
> void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
> diff --git a/mm/memblock.c b/mm/memblock.c
> index b68ee86788af..9277aca642b2 100644
> --- a/mm/memblock.c
> +++ b/mm/memblock.c
> @@ -1857,6 +1857,30 @@ void __init_memblock memblock_trim_memory(phys_addr_t
> align)
> }
> }
>
> +/**
> + * memblock_enforce_memory_reserved_overlap - make sure every range in
> + * @memblock.reserved is covered by @memblock.memory
> + *
> + * The data in @memblock.memory is used to detect zone and node boundaries
> + * during initialization of the memory map and the page allocator. Make
> + * sure that every memory range present in @memblock.reserved is also added
> + * to @memblock.memory even if the architecture specific memory
> + * initialization failed to do so
> + */
> +void __init memblock_enforce_memory_reserved_overlap(void)
> +{
> + phys_addr_t start, end;
> + int nid;
> + u64 i;
> +
> + __for_each_mem_range(i, &memblock.reserved, &memblock.memory,
> + NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, &nid) {
> + pr_warn("memblock: reserved range [%pa-%pa] is not in memory\n",
> + &start, &end);
> + memblock_add_node(start, (end - start), nid);
> + }
> +}
> +
> void __init_memblock memblock_set_current_limit(phys_addr_t limit)
> {
> memblock.current_limit = limit;
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index eaa227a479e4..dbc57dbbacd8 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -7436,6 +7436,13 @@ void __init free_area_init(unsigned long *max_zone_pfn)
> memset(arch_zone_highest_possible_pfn, 0,
> sizeof(arch_zone_highest_possible_pfn));
>
> + /*
> + * Some architectures (e.g. x86) have reserved pages outside of
> + * memblock.memory. Make sure these pages are taken into account
> + * when detecting zone and node boundaries
> + */
> + memblock_enforce_memory_reserved_overlap();
> +
> start_pfn = find_min_pfn_with_active_regions();
> descending = arch_has_descending_max_zone_pfns();
>
>
CCing Dan.
This implies that any memory that is E820_TYPE_SOFT_RESERVED that was
reserved via memblock_reserve() will be added via memblock_add_node() as
well, resulting in all such memory getting a memmap allocated right when
booting up, right?
IIRC, there are use cases where that is absolutely not desired.
Am I missing something? (@Dan?)
--
Thanks,
David / dhildenb
